I will not make any statements to the GPU-Z shot above, but I wanted to ask a general question: is that a believable standard versus boost clock ratio? 3.5%? I mean, what would be the point?
You are correct, the gap between base clock and boost seems smaller than usual.
Anything smaller than 5% is not worthwhile, other than to say "it has BOOST™".
Bouncing Zabaglione Bros.
Legend
Marketing purposes.
With NVIDIA GPUs, the advertised boost frequency is not the maximum clock speed the chip can reach, but the one you're pretty much guaranteed to see on a frequent basis with any sample.
In practice, most chips will probably go a few bins above that—provided the information is correct, of course—and it's not surprising that NVIDIA should promise a little less on a mid-range product.
It's not impossible. Nvidia has had similarly lopsided default configurations on at least three cards in the past couple generations. It's less expensive than going to the full 4GB and solves the 2GB VRAM longevity issue, at the cost of max bandwidth. This problem only comes into play when using more than 2GB, and can potentially be mitigated -- from my understanding at least. Nvidia hasn't revealed much about their implementation, however. The good news here, vs. the 192 bit busses using 1GB/2GB, is that using the extra address space should have less of a performance impact (1/2 the bandwidth vs. 1/3). In either case, 1/2 or 1/3 the bandwidth still ends up being higher than DDR3, and avoids the latency hit.
4GB is overkill; 3GB is future proof to the point that the GPU will be irrelevant before the VRAM is; 2GB is okay today, but will not be even potentially as soon as next year (On an unrelated note: 3 semicolons in one sentence, used correctly. Achievement unlocked! The odd things I pride myself in...). Many games already are cutting it close, and with the updated consoles inbound, frame buffers will undoubtedly require more VRAM.
FWIW this would be the highest clocked gk104, and it is not surprising that you get less boost on higher clocks (gtx770 which has clocks just barely below also has similar minimal guaranteed turbo).
No idea if it's legit or not but at 299$ it would perfectly fit I think. After all this will easily beat a gtx 670 (and a 7950 non-boost), about 15% slower than a gtx 770.
No idea if it's legit or not but at 299$ it would perfectly fit I think. After all this will easily beat a gtx 670 (and a 7950 non-boost), about 15% slower than a gtx 770.
I completely agree. The 760 Ti will certainly be faster than the 660 Ti. GK104's biggest bottleneck was with ROPs/bandwidth, rather than with texture and shader throughput. Disabling an SMX and reducing clocks to get the 670 only resulted in a ~5% reduction in performance compared to the 680.
The guy saying that it will be slower than the 660 Ti is completely misinformed. The shader and texture throughput with stock clocks will actually be higher on the 770 than the 670 (comparing Hz * shader count of each), although it will be interesting to see if the higher core count wins out over higher frequency, and I didn't get a chance to look at average boost clocks. However, the bandwidth on the 760 Ti will be markedly higher than the 670 and 660 Ti.
The 760 Ti should beat the 670, or be very, very close. The price reduction it offers should make it a very appealing card.
Edit: just checked the average boost clocks. When comparing Hz * shaders, the 670 has a tiny lead (3% higher). 3% is quite a bit in the context of this comparison, however the increased bandwidth on the 760 Ti should partially negate, fully negate, or potentially even pull ahead of that gap. If I remember correctly, the GPU Boost 2.0 also allows for the GPU to use higher boost states more frequently/longer.
Going to make a performance guess here: using TPU as my source, I'm expecting the 760 Ti to be on par with the 670, +3%/-2% (i.e., I believe it is more likely for the 760 Ti to be faster than the 670 than slower than it). This is using their "all resolutions" chart.
Last edited by a moderator:
Oh you're right I forgot to take the higher (relatively) boost on the 670 into account, so the 670 could indeed theoretically beat it by a small amount (and probably will in at least things like opencl). I think though the +17% memory bandwidth (and all around higher clock) should still win on average, but the difference to the 770 might be closer to 20%, the disabled GPC (?) might also make some difference possibly?
Also I guess on the OC front and perf/w it's probably not really all that great, but performance and price would still be right there where they belong compared to the 770 at least.
It has less cores than a 660ti, it is the OEM 660 with a 256bit bus.
It will not be faster than the 670.
$299 for this card would be overpriced.
It certainly can be faster the the 670, and most likely will be on par or slightly faster. The math supports this notion.
Even if it were slightly slower, it definitely would not be overpriced. Not even remotely close. It appears that AMDZone is leaking again...
http://wccftech.com/nvidia-geforce-gtx-760-overclocked-variant-performance-revealed-3dmark-11/
Slightly under a stock 670 score, that is including higher clocks for both core and memory...
Yes it would be overpriced at $299 because i can go buy a brand new 670 for $299 right now.
boxleitnerb
Regular
It clocks higher because it has fewer cores. What the market price will be, remains to be seen. What is the MSRP on the 670 currently? Certainly not $299...
3dmarks11, which mostly doesn't care about memory bandwidth and ROPs, just shaders.
I bet it will do slightly better on just about everything else, except similar shader-heavy stuff like opencl, but if you want fast performance there you better go buy a 7850...
Overpriced? Well maybe because the GTX 670 has slipped in price, this certainly isn't much of an upgrade. All I'm saying is in relation to the 399$ GTX 770 the price looks exactly what you'd expect (and not saying anything about retail prices just MSRP).
Ebay counts? Lol. Do you not know what an MSRP is?
3D Mark scores are heavily shader-favoring. The 680 has something like a 14% lead over the 670 in 3D Mark 11, but is only 7% faster on average. Coincidentally, the 670 has ever-so-slightly higher average shader throughput than the 760 Ti, which is perfectly reflected in its ever-so-slight lead over the 760 Ti.
The extra bandwidth will easily make up the difference.
Also, all previous rumors had placed this as the 760 Ti. The difference is wholly semantic. I highly doubt there will be a 760 Ti if this ends up being named the 760. As seen with the 670 vs. 680, disabling one SMX does not create a strong enough performance gap. The alternative, disabling 2 SMXs and clocking higher, is what they're doing here. It's an either/or deal.
Nvidia doesn't always use the Ti moniker. It would be perfectly sensible for them to skip it.
Last edited by a moderator:
http://www.centrecom.com.au/catalog/n670wf2-gigabyte-nvidia-geforce-n670wf2-p-60376.html?sort=2a
http://www.centrecom.com.au/catalog...670-ultra-durable-series-p-59790.html?sort=2a
Also in that bench the 760 is a factory overclocked version vs a stock 670.
The 670 will most likely have better overclocking potential, also it has better performance per clock.
Price of the 680 and 670 have been pulled down, certainly because they want empty the stock ...
Here you start find the 680 @ 300Euros ( -40-50Euros of initial price) and the 670 @ 260 Euros. ( -40, -50Euros ). Ok, thoses are special prices ( flash sold ) for the reason i have mention. then they will be EOL.
Here you start find the 680 @ 300Euros ( -40-50Euros of initial price) and the 670 @ 260 Euros. ( -40, -50Euros ). Ok, thoses are special prices ( flash sold ) for the reason i have mention. then they will be EOL.
Last edited by a moderator:
Occasionally they can be found for $300... but the vast majority are not selling for that. A good deal from most retailers on a 670 will run you about $340-350. Compare that to the vast majority of 760s selling for $300 or less... there's an obvious jump in bang for your buck. A handful of cards selling for $300 is not what will hurt AMD and boost Nvidia's sales -- there's not enough volume.
I didn't catch that those clocks were factory overclocked, though. In that case, the 670 has a better chance of winning out, but the edge will be negligible.
In response to overclocking, if the 760 has the same or a similar cooler as the 670, the 760 should easily overclock higher. The 700 series introduces more flexiblility with overvolting, and the memory ICs are better as well. Finally, they use a new revision of GK104. I'd imagine the 670 would do better when comparing max OC vs. max OC from a performance standpoint, but max clocks should easily be in favor of the 760 on average.
I didn't catch that those clocks were factory overclocked, though. In that case, the 670 has a better chance of winning out, but the edge will be negligible.
In response to overclocking, if the 760 has the same or a similar cooler as the 670, the 760 should easily overclock higher. The 700 series introduces more flexiblility with overvolting, and the memory ICs are better as well. Finally, they use a new revision of GK104. I'd imagine the 670 would do better when comparing max OC vs. max OC from a performance standpoint, but max clocks should easily be in favor of the 760 on average.
